I like this maths. Sorry if answered elsewhere, but are you working this right the way through the electoral college map to get to 2/3 Harris vs. 1/3 Trump?
Also, as a note to self, 33% for Trump is higher that the 25% chance of getting two heads in a row on consecutive coin flips, and getting two heads in a rows feel very plausible.
How much does “turnout amongst supporters” matter?
Say you have a 49-47 split in the polls. But 99% of 47s turn out, vs 95% of 49s. Undecideds split evenly.
My hunch is the Nates are over ascribing this “motivated voters” variance into their models, where trump got a lot of strong turnout from his supporters in the past, when in reality the trump overperformance was mostly due to undecideds not splitting evenly.
But wondering how much you consider this voter turnout factor in your model.
Turnout absolutely plays a role. As does motivation. The short answer is that it increases uncertainty. Trump's base will turn out. Trump's haters/Dem supporters will turn out. That's basically 45% for both candidates in every swing state, guaranteed. No ads needed.
The question becomes how we quantify that remaining 10%: not just swing voters, but passive ones.
You can see a little of that in "polls overestimated support" and "polls underestimated support"
Of course my forecast doesn't assume the poll average can't be wrong - but it does assume roughly equal probabilities that the polls have underestimated Harris/underestimated Trump.
Except nobody is really “undecided”. I’ve been out canvassing “undecided” for Harris and down ballot Democrats. I knock at the door and clearly identify myself with the Democrats, and people give themselves away.
Meticulously polite white men? Trump.
Friendly white men? Harris.
Careful white women? If they can get hold of their ballot away from the men in their family? Harris. If they can’t? Trump or maybe third party.
How would you characterize 2016 and 2020 in terms of the scenarios you outline above? The conventional answer I think is just that polls underestimated Trump in both years.
That's the conventional take but we know that he got disproportionate support from people who said they were undecided, and in 2016, people who said they favored a third-party also went for Trump
A good analyst would account for these factors... current methods simply assume that undecideds split 50 50 and no one changes their mind...even when that's clearly not true
Do you think the professionals who run campaigns see things your way or the more conventional and wrong way? I would imagine that if I am running someone's campaign, and I think of polls as "base of support," I am going to do things differently than if I think of them as a prediction.
A couple of years ago, David Plouffe did some interviews with campaign spokesmen in various swing state races, and he was always asking them how many votes they thought they needed to win and what was their plan to get there, so he, at least, seems like someone who might think in more "base of support" terms.
The tactical advantage to the "right way" and the "conventional way" are pretty narrow in most cases
After all, whether a campaign believes they have a 35% chance to win a state or a 65% chance won't impact their strategy much - it's still a big investment.
The exception would be a case like 2016 wherein Hillary was "up by 6" in certain states and that was perceived as a safe lead...when in reality, she was only around 46% in the poll averages - which indicates it's far from safe.
Overall, I would say campaigns have a better grasp on this stuff than the average person (and maybe even better than me - I can't say for certain) but given the impact of political races and the finite amount of time and money ..I would say there are some edges to be gained from a better understanding of polls, in any election - not just at the Presidential level, but for lower races as well
Makes sense. I guess what I was thinking is whether all campaigns are laser-focused on unducided voters or whether a "polls are base of support" candidate would be more focused and effective at brigning over undecideds than a "polls are predictions" candidate.
I guess a 45% or 46% candidate should be focused on continuing to build their base rather than getting a false sense of security from any margin they may have.
But, in, say 2022 OH Senate, a lot of those undecided voters that led to Ryan having a margin in the polls may not have been really attainable by the best of campaigns.
This is only popular vote. You’d need to do this state by state in the battlegrounds. As you did, starting from polling averages in each of the 7, her margin is smaller than national vote. It’s not that cut and dry as you suggest.
In scenario 1C, I do not know how you figure that Harris wins narrowly. That seems like a probable Trump victory in that scenario to me.
With the variability of polls, I don't see how drilling down too deeply into the tea leaves helps.
This is a very close race, and the polls consistently indicate that.
Turnout will be important.
I put the chances at something close to 50:50. The margins are too close to have much confidence in the precise outcome of the polls, and anecdotally, there are lots of factors in flux.
I make no prediction as to who will win the electoral college, and I think your analysis fails to adequately capture how much the social and political situation has changed from 2020. This is a different world. I could explain all the ways in which past projections are not valid assumptions now, but that's outside the scope of this.
Suffice it to say, to my analysis, variance high enough to encompass the tipping point in the state polling data, and the (in some ways) lack of any historical precedent for this situation, means that there is little confidence one can have in anything right now.
So it sounds like 2016 was a Scenario 1C, but what about 2020? Would you also consider that to be a 1C (since both candidates hit or exceeded their polling averages), or some type of Scenario 3?
I looked through the actual polls in Texas back to the start of September. The ones that show the data broken out into cross-sections usually shows a very low %age of respondents in the 18-24 (even up to mid 30s) age demographic, which leans heavily towards Harris. Is this an accurate reflection of the electorate or a polling flaw?
This is great, but being logical and pragmatic has no place in political commentary!
Do you still have Harris at 68%?
Yep, not much has changed. Down a few tenths of a percent in the polling averages, win probability still over 2/3
I like this maths. Sorry if answered elsewhere, but are you working this right the way through the electoral college map to get to 2/3 Harris vs. 1/3 Trump?
Also, as a note to self, 33% for Trump is higher that the 25% chance of getting two heads in a row on consecutive coin flips, and getting two heads in a rows feel very plausible.
How much does “turnout amongst supporters” matter?
Say you have a 49-47 split in the polls. But 99% of 47s turn out, vs 95% of 49s. Undecideds split evenly.
My hunch is the Nates are over ascribing this “motivated voters” variance into their models, where trump got a lot of strong turnout from his supporters in the past, when in reality the trump overperformance was mostly due to undecideds not splitting evenly.
But wondering how much you consider this voter turnout factor in your model.
Turnout absolutely plays a role. As does motivation. The short answer is that it increases uncertainty. Trump's base will turn out. Trump's haters/Dem supporters will turn out. That's basically 45% for both candidates in every swing state, guaranteed. No ads needed.
The question becomes how we quantify that remaining 10%: not just swing voters, but passive ones.
You can see a little of that in "polls overestimated support" and "polls underestimated support"
Of course my forecast doesn't assume the poll average can't be wrong - but it does assume roughly equal probabilities that the polls have underestimated Harris/underestimated Trump.
Except nobody is really “undecided”. I’ve been out canvassing “undecided” for Harris and down ballot Democrats. I knock at the door and clearly identify myself with the Democrats, and people give themselves away.
Meticulously polite white men? Trump.
Friendly white men? Harris.
Careful white women? If they can get hold of their ballot away from the men in their family? Harris. If they can’t? Trump or maybe third party.
Kamala harris has like a 90% chance of winning the popular vote, no?
Yeah give or take
How would you characterize 2016 and 2020 in terms of the scenarios you outline above? The conventional answer I think is just that polls underestimated Trump in both years.
That's the conventional take but we know that he got disproportionate support from people who said they were undecided, and in 2016, people who said they favored a third-party also went for Trump
A good analyst would account for these factors... current methods simply assume that undecideds split 50 50 and no one changes their mind...even when that's clearly not true
Talk about it in more depth in my book :)
Do you think the professionals who run campaigns see things your way or the more conventional and wrong way? I would imagine that if I am running someone's campaign, and I think of polls as "base of support," I am going to do things differently than if I think of them as a prediction.
A couple of years ago, David Plouffe did some interviews with campaign spokesmen in various swing state races, and he was always asking them how many votes they thought they needed to win and what was their plan to get there, so he, at least, seems like someone who might think in more "base of support" terms.
The tactical advantage to the "right way" and the "conventional way" are pretty narrow in most cases
After all, whether a campaign believes they have a 35% chance to win a state or a 65% chance won't impact their strategy much - it's still a big investment.
The exception would be a case like 2016 wherein Hillary was "up by 6" in certain states and that was perceived as a safe lead...when in reality, she was only around 46% in the poll averages - which indicates it's far from safe.
Overall, I would say campaigns have a better grasp on this stuff than the average person (and maybe even better than me - I can't say for certain) but given the impact of political races and the finite amount of time and money ..I would say there are some edges to be gained from a better understanding of polls, in any election - not just at the Presidential level, but for lower races as well
Makes sense. I guess what I was thinking is whether all campaigns are laser-focused on unducided voters or whether a "polls are base of support" candidate would be more focused and effective at brigning over undecideds than a "polls are predictions" candidate.
I guess a 45% or 46% candidate should be focused on continuing to build their base rather than getting a false sense of security from any margin they may have.
But, in, say 2022 OH Senate, a lot of those undecided voters that led to Ryan having a margin in the polls may not have been really attainable by the best of campaigns.
This is only popular vote. You’d need to do this state by state in the battlegrounds. As you did, starting from polling averages in each of the 7, her margin is smaller than national vote. It’s not that cut and dry as you suggest.
Yeah!!!
I know that there are a lot of partisan polling operations spinning for Trump.
Is Rasmussen one of them? Or are they "legit"?
I really don't know. They just seem link an "out an out liar" (I think that's a statistical term.)
In scenario 1C, I do not know how you figure that Harris wins narrowly. That seems like a probable Trump victory in that scenario to me.
With the variability of polls, I don't see how drilling down too deeply into the tea leaves helps.
This is a very close race, and the polls consistently indicate that.
Turnout will be important.
I put the chances at something close to 50:50. The margins are too close to have much confidence in the precise outcome of the polls, and anecdotally, there are lots of factors in flux.
I make no prediction as to who will win the electoral college, and I think your analysis fails to adequately capture how much the social and political situation has changed from 2020. This is a different world. I could explain all the ways in which past projections are not valid assumptions now, but that's outside the scope of this.
Suffice it to say, to my analysis, variance high enough to encompass the tipping point in the state polling data, and the (in some ways) lack of any historical precedent for this situation, means that there is little confidence one can have in anything right now.
So it sounds like 2016 was a Scenario 1C, but what about 2020? Would you also consider that to be a 1C (since both candidates hit or exceeded their polling averages), or some type of Scenario 3?
I looked through the actual polls in Texas back to the start of September. The ones that show the data broken out into cross-sections usually shows a very low %age of respondents in the 18-24 (even up to mid 30s) age demographic, which leans heavily towards Harris. Is this an accurate reflection of the electorate or a polling flaw?
Why did Cook start changing Lean Ds today to toss ups?
In my observation they move things based on the poll margin
Polls/poll average of 0-2 "tossup"
3-6 "lean"
6-12 "likely"
12+ "solid"
Give or take of course